{"id":2168,"date":"2024-03-21T15:00:30","date_gmt":"2024-03-21T15:00:30","guid":{"rendered":"https:\/\/cog-ist.com\/?post_type=blog_content&p=2168"},"modified":"2025-09-20T14:17:18","modified_gmt":"2025-09-20T14:17:18","slug":"kara-kutuyu-yorumlamak-buyuk-dil-modelleri-ve-dil-bilgisi-bicimbilim-deniz-ekin-yavas","status":"publish","type":"blog_content","link":"https:\/\/cog-ist.com\/en\/blog_content\/kara-kutuyu-yorumlamak-buyuk-dil-modelleri-ve-dil-bilgisi-bicimbilim-deniz-ekin-yavas\/","title":{"rendered":"Kara Kutuyu Yorumlamak: B\u00fcy\u00fck Dil Modelleri ve Dil Bilgisi \u2014 Bi\u00e7imbilim \u2014 Deniz Ekin Yava\u015f"},"content":{"rendered":"\n

\u201cKara Kutuyu Yorumlamak\u201d serisinin t\u00fcm yaz\u0131lar\u0131na buradan<\/a> eri\u015febilirsiniz.<\/p>\n\n\n\n

Deniz Ekin Yava\u015f, Heinrich-Heine \u00dcniversitesi\u2019nde Hesaplamal\u0131 Dilbilim alan\u0131nda doktora \u00f6\u011frencisi ve ara\u015ft\u0131rma g\u00f6revlisi. \u00d6ne\u011fitimli dil modellerini kullanarak s\u00f6zl\u00fcksel anlambilim ve anlambilim-s\u00f6zdizim kesi\u015fimini ara\u015ft\u0131r\u0131yor.<\/em><\/p>\n\n\n\n

Son y\u0131llarda b\u00fcy\u00fck dil modelleri bir\u00e7ok farkl\u0131 g\u00f6revde elde ettikleri ba\u015far\u0131larla b\u00fcy\u00fck ilgi \u00e7ekti. Bu ba\u015far\u0131lar, bizlere b\u00fcy\u00fck dil modellerinin \u00f6ne\u011fitim (pre-training)<\/em> s\u00fcre\u00e7lerinde dilin istatistiksel bilgisinin \u00f6tesinde dile ili\u015fkin \u00e7ok daha fazlas\u0131n\u0131 \u00f6\u011frenip \u00f6\u011frenmedi\u011fini sormam\u0131za neden oldu. Bunun sonucu olarak bu soruyu yan\u0131tlamay\u0131 ama\u00e7layan yorumlanabilirlik (interpretability)<\/em> \u00e7al\u0131\u015fmalar\u0131 bilimsel alanda b\u00fcy\u00fck bir \u00f6nem kazand\u0131. Bu serinin amac\u0131, farkl\u0131 konulardaki yorumlanabilirlik \u00e7al\u0131\u015fmalar\u0131na de\u011finerek modellerin bili\u015fsel olarak ger\u00e7ek\u00e7ili\u011fini (cognitive plausibility)<\/em> sorgulamak.<\/p>\n\n\n\n

Bu yaz\u0131 serisinde ge\u00e7en yaz\u0131yla beraber oda\u011f\u0131m\u0131z\u0131 dilbilgisine \u00e7ekmi\u015f ve s\u00f6zdizimine odaklanm\u0131\u015ft\u0131k. Bu yaz\u0131da ise bi\u00e7imbilime (morphology)<\/em> de\u011finece\u011fiz. Bi\u00e7imbilim, bir dildeki s\u00f6zc\u00fcklerin i\u00e7 yap\u0131s\u0131n\u0131 inceleyen dilbilimin alt dal\u0131d\u0131r. Bi\u00e7imbilim bir dilde yeni s\u00f6zc\u00fcklerin nas\u0131l olu\u015ftu\u011funu ve s\u00f6zc\u00fcklerin \u00e7ekimlenerek t\u00fcmcelerde belirli dilbilgisel i\u015flevlerini nas\u0131l yerine getirdiklerini inceler.<\/p>\n\n\n\n

Bu yaz\u0131da b\u00fcy\u00fck dil modellerinin bi\u00e7imbilime ili\u015fkin bilgi edinip edinmedi\u011fini sorgulayaca\u011f\u0131z ve bu konuda yap\u0131lm\u0131\u015f yorumlanabilirlik \u00e7al\u0131\u015fmalar\u0131na yer verece\u011fiz. Bu \u00e7al\u0131\u015fmalara ge\u00e7meden \u00f6nce bi\u00e7imbilime ve temel kavramlar\u0131na de\u011finerek ba\u015flayal\u0131m.<\/p>\n\n\n\n

Bi\u00e7imbilim ve Temel Kavramlar\u0131<\/strong><\/p>\n\n\n\n

Bi\u00e7imbilim, bir dildeki s\u00f6zc\u00fcklerin i\u00e7 yap\u0131s\u0131n\u0131 inceleyen dilbilimin alt dal\u0131d\u0131r. S\u00f6zc\u00fckler bi\u00e7imbirimlerden (morpheme) <\/em>olu\u015fur. Bir dilin bi\u00e7imbirimleri asl\u0131nda o dildeki k\u00f6kler (roots)<\/em> ve eklere (aff\u0131xes) <\/em>denk gelmektedir. Bir s\u00f6zc\u00fck bir bi\u00e7imbirimden olu\u015fabilece\u011fi gibi birden \u00e7ok bi\u00e7imbirimin bir araya gelmesiyle de olu\u015fabilir. \u00d6rne\u011fin, \u2018kitap\u2019 s\u00f6zc\u00fc\u011f\u00fc tek bir bi\u00e7imbirimden (\u2018kitap\u2019 k\u00f6k\u00fc) olu\u015furken, \u2018kitapl\u0131k\u2019 s\u00f6zc\u00fc\u011f\u00fc iki bi\u00e7imbirimden olu\u015fur (\u2018kitap\u2019 k\u00f6k\u00fc ve \u2018-l\u0131k\u2019 eki). Sadece bir bi\u00e7imbirimden olu\u015fan s\u00f6zc\u00fcklere basit s\u00f6zc\u00fck (simple word)<\/em>, birden \u00e7ok bi\u00e7imbirimden olu\u015fan s\u00f6zc\u00fcklere ise karma\u015f\u0131k s\u00f6zc\u00fck (complex word)<\/em> ad\u0131 verilir.<\/p>\n\n\n\n

Eklerin bir i\u015flevi yeni s\u00f6zc\u00fck \u00fcretimidir. Bi\u00e7imbilimde buna t\u00fcretim (derivation)<\/em> ad\u0131 verilir. Ekler ba\u015fka bi\u00e7imbirimlerle birle\u015ferek yeni s\u00f6zc\u00fckler \u00fcretir. Ekler basit s\u00f6zc\u00fcklere eklenerek karma\u015f\u0131k s\u00f6zc\u00fckler olu\u015fturabilir, \u2018kitapl\u0131k\u2019 (\u2018kitap\u2019, \u2018-l\u0131k\u2019) \u00f6rne\u011finde oldu\u011fu gibi ya da halihaz\u0131rda karma\u015f\u0131k s\u00f6zc\u00fcklere eklenerek ba\u015fka karma\u015f\u0131k s\u00f6zc\u00fckler olu\u015fturabilir, \u00f6rne\u011fin \u2018g\u00f6zl\u00fck\u2019 (\u2018g\u00f6z\u2019, \u2018-l\u00fck\u2019) s\u00f6zc\u00fc\u011f\u00fcnden \u2018g\u00f6zl\u00fck\u00e7\u00fc\u2019 (\u2018g\u00f6z\u2019, \u2018-l\u00fck\u2019, \u2018-\u00e7\u00fc\u2019) s\u00f6zc\u00fc\u011f\u00fcn\u00fcn t\u00fcretilmesinde oldu\u011fu gibi.<\/p>\n\n\n\n

Eklerin ba\u015fka \u00f6nemli bir i\u015flevi ise s\u00f6zdizimiyle, yani s\u00f6zc\u00fcklerin t\u00fcmcelerdeki dilgisel i\u015fleviyle yak\u0131ndan ili\u015fkilidir. Bu ama\u00e7la ekler belirli bir dilbilgisel bilgiyi ifade etmek i\u00e7in s\u00f6zc\u00fcklere eklenir. Bi\u00e7imbilimde buna \u00e7ekim (inflection)<\/em> ad\u0131 verilir. \u00c7ekime \u00f6rnek olarak adlar i\u00e7in durum (case)<\/em>, say\u0131 (number)<\/em>, dilbilgisel cinsiyet (gender)<\/em> \u00e7ekimleri; eylemler i\u00e7in zaman (tense)<\/em>, ki\u015fi (person)<\/em> \u00e7ekimleri verilebilir. Farkl\u0131 s\u00f6zc\u00fck t\u00fcrleri i\u00e7in \u00e7ekim \u00f6rnekleri a\u015fa\u011f\u0131da g\u00f6r\u00fclebilir. Her dil ayn\u0131 \u00e7ekim \u00f6zelliklerine sahip de\u011fildir. \u00d6rne\u011fin T\u00fcrk\u00e7e\u2019de adlar dilbilgisel cinsiyetlerine g\u00f6re \u00e7ekimlenmez ya da benzer bir \u015fekilde \u0130ngilizce\u2019de adlar duruma g\u00f6re \u00e7ekimlenmez.<\/p>\n\n\n\n

    \n
  1. Adlarda say\u0131:\u00a0<\/strong>T\u00fcrk\u00e7e\u2019de tekil (\u2018kitap\u2019), \u00e7o\u011ful (\u2018kitaplar\u2019)<\/li>\n\n\n\n
  2. Adlarda durum:\u00a0<\/strong>T\u00fcrk\u00e7e\u2019de yal\u0131n (\u2018kitap\u2019), belirtme (\u2018kitab\u0131\u2019), y\u00f6nelme (\u2018kitaba\u2019), \u2026<\/li>\n\n\n\n
  3. Adlarda dilbilgisel cinsiyet:<\/strong>\u00a0Almanca\u2019da eril (\u2018der Arzt\u2019), di\u015fil (\u2018die \u00c4rztin\u2019), \u2026<\/li>\n\n\n\n
  4. Eylemlerde zaman:\u00a0<\/strong>T\u00fcrk\u00e7e\u2019de gelecek (\u2018okuyacak\u2019), ge\u00e7mi\u015f (\u2018okudu\u2019), \u2026<\/li>\n\n\n\n
  5. Eylemlerde ki\u015fi ve say\u0131:\u00a0<\/strong>T\u00fcrk\u00e7e\u2019de birinci tekil (\u2018ald\u0131m\u2019), birinci \u00e7o\u011ful (\u2018ald\u0131k\u2019), \u2026<\/li>\n<\/ol>\n\n\n\n

    Bi\u00e7imbilim \u00d6zelinde Yorumlanabilirlik \u00c7al\u0131\u015fmalar\u0131<\/strong><\/p>\n\n\n\n

    Bu yaz\u0131da ele alaca\u011f\u0131m\u0131z \u00e7al\u0131\u015fmalar daha \u00f6nceki yaz\u0131lar\u0131m\u0131zda de\u011findi\u011fimiz farkl\u0131 yorumlanabilirlik y\u00f6ntemlerini kullanmaktad\u0131r. Bunlardan bir tanesi sorgulay\u0131c\u0131 s\u0131n\u0131fland\u0131r\u0131c\u0131lard\u0131r (probing classifiers)<\/em>. Sorgulay\u0131c\u0131 s\u0131n\u0131fland\u0131r\u0131c\u0131lar\u0131n kullan\u0131m amac\u0131 b\u00fcy\u00fck dil modellerinin s\u00f6zc\u00fck g\u00f6mmelerindeki bilgiyi ortaya \u00e7\u0131karmakt\u0131r. Bu s\u0131n\u0131fland\u0131r\u0131c\u0131larla s\u00f6zc\u00fck g\u00f6mmeleri, dilbilgisel \u00f6zelliklerine g\u00f6re s\u0131n\u0131fland\u0131r\u0131l\u0131r. Bunun i\u00e7in s\u0131n\u0131fland\u0131r\u0131c\u0131ya bir s\u00f6zc\u00fck g\u00f6mmesi girdi olarak verilir ve s\u0131n\u0131fland\u0131r\u0131c\u0131 g\u00f6mmeden yola \u00e7\u0131karak bu g\u00f6mmenin temsil etti\u011fi s\u00f6zc\u00fc\u011f\u00fcn dilbilgisel \u00f6zelliklerini tahmin etmeye \u00e7al\u0131\u015f\u0131r. Bu s\u0131n\u0131fland\u0131r\u0131c\u0131n\u0131n performans\u0131 bize bu bilginin ne kadar ba\u015far\u0131l\u0131 bir \u015fekilde s\u00f6zc\u00fck g\u00f6mmesinden \u00e7\u0131kar\u0131labildi\u011fini g\u00f6sterir.<\/p>\n\n\n\n

    Acs vd. (2023) \u00e7al\u0131\u015fmalar\u0131nda \u00e7okdilli b\u00fcy\u00fck dil modellerinin (multilingual large language models)<\/em> bi\u00e7imbilime ili\u015fkin bilgilerini test etmeyi ama\u00e7lar. \u00c7okdilli modeller, birden fazla dil verisi ile e\u011fitilmi\u015f modellerdir ve e\u011fitildikleri b\u00fct\u00fcn dillere ili\u015fkin bilgileri \u00f6\u011frenirler. Bu t\u00fcr modellerden en pop\u00fclerleri BERT modelinin ve RoBERTa modelinin \u00e7okdilli versiyonlar\u0131 mBERT ve XLM-RoBERTa\u2019d\u0131r. Bu iki model de yakla\u015f\u0131k 100 dilin verisiyle e\u011fitilmi\u015ftir.<\/p>\n\n\n\n

    Acs vd. (2023), mBERT ve XLM-RoBERTa\u2019n\u0131n bi\u00e7imbilim bilgisini test etmek i\u00e7in 10 dil ailesinden 42 dil i\u00e7in toplam 247 sorgulama testi geli\u015ftirir. Bu testlerle s\u0131fat, ad, eylem ve ad\u0131llar\u0131n farkl\u0131 \u00e7ekimsel \u00f6zelliklerini test ederler, \u00f6rne\u011fin, adlar i\u00e7in durum, eylemler i\u00e7in ki\u015fi ve zaman, vb. Bu ama\u00e7la modellerin s\u00f6zc\u00fck g\u00f6mmeleri ile her bilgi t\u00fcr\u00fc i\u00e7in farkl\u0131 s\u0131n\u0131fland\u0131r\u0131c\u0131lar e\u011fitilir. \u00d6rne\u011fin, G\u00f6rsel 1\u2019deki gibi adlar\u0131n say\u0131 \u00f6zellikleri i\u00e7in e\u011fitilen bir s\u0131n\u0131fland\u0131r\u0131c\u0131dan bir ad\u0131n tekil mi yoksa \u00e7o\u011ful mu oldu\u011funu tahmin etmesi beklenir.<\/p>\n\n\n\n

    \"\"\/
    G\u00f6rsel 1: \u2018elma\u2019 ve \u2018elmalar\u2019 s\u00f6zc\u00fcklerinin say\u0131 \u00f6zelliklerine g\u00f6re s\u0131n\u0131fland\u0131r\u0131lmas\u0131<\/figcaption><\/figure>\n\n\n\n

    Bu iki \u00e7okdilli b\u00fcy\u00fck dil modelinin s\u00f6zc\u00fck g\u00f6mmelerinin performans\u0131, FastText statik s\u00f6zc\u00fck g\u00f6mmeleri ve Stanza ara\u00e7 seti (toolkit)<\/em>[1]<\/em><\/strong><\/a> <\/em>ile kar\u015f\u0131la\u015ft\u0131r\u0131l\u0131r. Stanza, bir\u00e7ok dil i\u00e7in olduk\u00e7a ba\u015far\u0131l\u0131 do\u011fal dil i\u015fleme ara\u00e7lar\u0131 ve \u00f6ne\u011fitimli modeller sunar. Testlerin sonucunda \u00e7okdilli b\u00fcy\u00fck dil modellerinin statik s\u00f6zc\u00fck g\u00f6mmelerini a\u00e7\u0131k ara farkla ge\u00e7ti\u011fi ve Stanza modellerine yak\u0131n performans g\u00f6sterdi\u011fi ortaya \u00e7\u0131kmaktad\u0131r. Bu \u00f6nemli bir bulgudur \u00e7\u00fcnk\u00fc Stanza modelleri etiketlenmi\u015f dil veriyle, farkl\u0131 diller i\u00e7in \u00f6zellikle bu ama\u00e7la e\u011fitilmi\u015ftir. \u00c7okdilli b\u00fcy\u00fck dil modelleri ise sadece dillere maruz kalarak \u00f6\u011frendikleri bilgi ile Stanza modellerine yak\u0131n performans g\u00f6sterebilmektedir.<\/p>\n\n\n\n

    Bu yaz\u0131da ele alaca\u011f\u0131m\u0131z di\u011fer iki \u00e7al\u0131\u015fma ise y\u00f6ntemsel olarak istem <\/em>(prompt<\/em>) temelli tan\u0131 testlerini kullanmaktad\u0131r. \u0130stem temelli testlerde, modele bir istem sunulur ve modelden bu istem do\u011frultusunda s\u00f6zc\u00fck tahmini yapmas\u0131 beklenir. Ara\u015ft\u0131rmac\u0131lar, istem temelli testler arac\u0131l\u0131\u011f\u0131yla modellerin belirli dilsel ba\u011flamlardaki davran\u0131\u015flar\u0131n\u0131 incelerler.<\/p>\n\n\n\n

    Hoffmann vd. (2020) istem temelli tan\u0131 testi ile \u0130ngilizce BERT modelinin s\u00f6zc\u00fck t\u00fcretimine ili\u015fkin bilgisini ortaya \u00e7\u0131karmay\u0131 ama\u00e7lar. BERT modeline bir t\u00fcmce ve bir s\u00f6zc\u00fck verildi\u011finde modelin, o s\u00f6zc\u00fckten o t\u00fcmce i\u00e7in uygun bir s\u00f6zc\u00fck t\u00fcretip t\u00fcretemedi\u011fini ara\u015ft\u0131r\u0131rlar. \u00d6rne\u011fin, G\u00f6rsel 2\u2019deki gibi, modele \u2018Bu yemek \u00e7ok doyur__.\u2019 t\u00fcmcesi verildi\u011finde modelden bo\u015fluk yerine, \u2018doyur-\u2019 s\u00f6zc\u00fc\u011f\u00fc i\u00e7in uygun ekin \u00fcretilmesi beklenir. Bu t\u00fcmce i\u00e7in do\u011fru t\u00fcretim ise \u2018doyurucu\u2019 olacakt\u0131r. Ara\u015ft\u0131rman\u0131n sonu\u00e7lar\u0131, modelin sadece testin %30\u2019u i\u00e7in do\u011fru yan\u0131t\u0131 \u00fcretebildi\u011fini g\u00f6stermektedir.<\/p>\n\n\n\n

    \"\"\/
    G\u00f6rsel 2: BERT modelinin t\u00fcretim eki tahmini<\/figcaption><\/figure>\n\n\n\n

    Haley (2020) ise istem temelli tan\u0131 testi y\u00f6ntemi ile b\u00fcy\u00fck dil modellerinin \u00e7ekim bilgisinin genellebilirli\u011fini ara\u015ft\u0131r\u0131r. Haley (2020) \u00e7al\u0131\u015fmas\u0131nda insanlar\u0131n bi\u00e7imbilimsel genelleme kapasitesini \u00f6l\u00e7en Wug Testi\u2019nden (Berko, 1958) esinlenmi\u015ftir. Bu testte konu\u015fuculara bilmedikleri bir s\u00f6zc\u00fck bir t\u00fcmce i\u00e7erisinde istem olarak verilir ve bu s\u00f6zc\u00fc\u011f\u00fc i\u00e7eren ba\u015fka bir t\u00fcmceyi tamamlamalar\u0131 istenir. Bu ikinci t\u00fcmce, s\u00f6zc\u00fc\u011f\u00fcn belirli bir formda tamamlanmas\u0131 i\u00e7in tasarlanm\u0131\u015ft\u0131r. G\u00f6rsel 3\u2019de bilinmeyen s\u00f6zc\u00fc\u011f\u00fcn \u00e7o\u011ful ekiyle tamamlanmas\u0131 i\u00e7in tasarlanan Wug Testi \u00f6rne\u011fi verilmi\u015ftir. (\u201cBu bir wug. \u015eimdi bir tane daha var. Onlardan iki tane var. \u0130ki tane ___ var.\u201d)<\/p>\n\n\n\n

    \"\"\/
    G\u00f6rsel 3: Say\u0131 \u00e7ekiminin genellenebilirli\u011finin testi i\u00e7in tasarlanm\u0131\u015f Wug Testi<\/figcaption><\/figure>\n\n\n\n

    Haley (2020) \u00e7al\u0131\u015fmas\u0131nda farkl\u0131 diller i\u00e7in e\u011fitilmi\u015f BERT modellerinin (Almanca, \u0130ngilizce, Frans\u0131zca, Felemenk\u00e7e ve \u0130spanyolca) \u00f6zne-eylem uyumuna ili\u015fkin bilgilerinin genellebilirli\u011fini \u00f6l\u00e7er. \u00d6zne-eylem uyumu, bir c\u00fcmlenin \u00f6znesi ile eylemi aras\u0131ndaki ki\u015fi ve say\u0131 a\u00e7\u0131s\u0131ndan uyumu ifade eder. \u00d6rne\u011fin, \u201cBen geliyorum.\u201d c\u00fcmlesinde \u00f6zne olan \u2018ben\u2019 birinci tekil ki\u015fidir ve buna uygun olarak eylemin \u00e7ekimli hali olan \u2018geliyorum\u2019 da birinci tekil ki\u015fiye g\u00f6re \u00e7ekilmi\u015ftir. Daha \u00f6nceki \u00e7al\u0131\u015fmalar BERT modelinin bu konuda bilgi sahibi oldu\u011funu g\u00f6stermektedir (Goldberg, 2019). Haley (2020) ise bu bilginin genellebilirli\u011fini \u00f6l\u00e7meyi ama\u00e7lar.<\/p>\n\n\n\n

    Bu ama\u00e7la Wug Testi\u2019nden yola \u00e7\u0131karak bu diller i\u00e7in istem temelli tan\u0131 testleri tasarlar. Bu testlerde modele istem olarak o dilde var olmayan bir s\u00f6zc\u00fc\u011f\u00fc i\u00e7eren basit bir t\u00fcmce verilir (\u2018This is a bik\u2019, \u2018Bu bir bik.\u2019). Modelden bu t\u00fcmceye kar\u015f\u0131l\u0131k ikinci t\u00fcmcede o s\u00f6zc\u00fc\u011f\u00fcn \u00e7o\u011ful veya tekil halleri i\u00e7in uyumlu eylemler \u00fcretmesi beklenir (\u2018The bik laughs\u2019\/\u2018The biks laugh\u2019, \u2018Bik g\u00fcler.\u2019\/\u2018Bikler g\u00fclerler.\u2019). Bu testi ayn\u0131 zamanda dildeki var olan s\u00f6zc\u00fckler i\u00e7in de yapar. Bu iki durumun kar\u015f\u0131la\u015ft\u0131r\u0131lmas\u0131 bize modelin veriyi ezberleme yetene\u011finin \u00f6tesinde \u00f6zne-y\u00fcklem uyumuna ili\u015fkin \u00e7ekim \u00f6zelliklerini bilip bilmedi\u011fini, yani bu bilginin genellenebilirli\u011fini g\u00f6sterir.<\/p>\n\n\n\n

    \"\"\/
    G\u00f6rsel 4: BERT i\u00e7in tasarlanm\u0131\u015f \u00f6zne-eylem uyumunun genellenebilirli\u011finin testi.<\/figcaption><\/figure>\n\n\n\n

    \u00c7al\u0131\u015fman\u0131n sonu\u00e7lar\u0131, modelin hem dilde var olan hem de var olmayan s\u00f6zc\u00fckler i\u00e7in \u00e7o\u011funlukla do\u011fru formu \u00fcretti\u011fini g\u00f6stermektedir. Bu da modellerin \u00f6zne-y\u00fcklem uyumuna ili\u015fkin \u00e7ekim \u00f6zelliklerini genelleyebildiklerini g\u00f6sterir.<\/p>\n\n\n\n

    Sonu\u00e7<\/strong><\/p>\n\n\n\n

    Serinin bu yaz\u0131s\u0131nda b\u00fcy\u00fck dil modellerinin bi\u00e7imbilime ili\u015fkin bilgilerini sorgulayan yorumlanabilirlik \u00e7al\u0131\u015fmalar\u0131na de\u011findik. Bu \u00e7al\u0131\u015fmalar\u0131n sonucunda hem tek dilli hem de \u00e7okdilli b\u00fcy\u00fck dil modellerinin s\u00f6zc\u00fcklerin \u00e7ekimsel \u00f6zelliklerine ili\u015fkin bilgiyi s\u00f6zc\u00fck g\u00f6mmelerinde kodlad\u0131\u011f\u0131n\u0131 ve bu bilgiyi genelleyebildiklerini ancak t\u00fcretimin bu modeller i\u00e7in zor oldu\u011funu g\u00f6rd\u00fck.<\/p>\n\n\n\n

    Notlar<\/strong><\/p>\n\n\n\n

    [1]<\/a> https:\/\/stanfordnlp.github.io\/stanza\/<\/a><\/p>\n\n\n\n

    Kaynak\u00e7a<\/strong><\/p>\n\n\n\n

    Acs, J., Hamerlik, E., Schwartz, R., Smith, N. A., & Kornai, A. (2023). Morphosyntactic probing of multilingual BERT models. Natural Language Engineering, 1\u201340.<\/p>\n\n\n\n

    Berko, J. (1958). The child\u2019s learning of English morphology. Word, 14(2\u20133), 150\u2013177.<\/p>\n\n\n\n

    Goldberg, Y. (2019). Assessing BERT\u2019s syntactic abilities. arXiv preprint arXiv:1901.05287.<\/p>\n\n\n\n

    Haley, C. (2020). This is a BERT. Now there are several of them. Can they generalize to novel words?. In Proceedings of the Third BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP (pp. 333\u2013341).<\/p>\n\n\n\n

    Hofmann, V., Pierrehumbert, J. B., & Sch\u00fctze, H. (2020). DagoBERT: Generating derivational morphology with a pretrained language model. arXiv preprint arXiv:2005.00672.<\/p>\n","protected":false},"featured_media":2169,"template":"","meta":{"_acf_changed":false},"event_publishing_tags":[479,426,84,280,286,94,93,924,627,342,416,345,691,64,923,92,323,281,285,87,1057,62,1056,348,88,1058,1059,1053,341,89,343,344,76,1054,1055,346,425,61,349,96],"kategori":[725],"class_list":["post-2168","blog_content","type-blog_content","status-publish","has-post-thumbnail","hentry","event_publishing_tags-ai","event_publishing_tags-artificial","event_publishing_tags-artificial-intelligence","event_publishing_tags-bilgisayar","event_publishing_tags-bilgisayar-bilimi","event_publishing_tags-bilis","event_publishing_tags-bilissel-bilim","event_publishing_tags-bilissel-dilbilim","event_publishing_tags-buyuk-dil-modeli","event_publishing_tags-buyuk-dil-modelleri","event_publishing_tags-chatbot","event_publishing_tags-chatgpt","event_publishing_tags-cogist","event_publishing_tags-cognition","event_publishing_tags-cognitive-linguistics","event_publishing_tags-cognitive-science","event_publishing_tags-cogsci","event_publishing_tags-computer","event_publishing_tags-computer-science","event_publishing_tags-dil","event_publishing_tags-dil-modeli","event_publishing_tags-dilbilim","event_publishing_tags-dogal-dil-isleme","event_publishing_tags-intelligence","event_publishing_tags-language","event_publishing_tags-language-model","event_publishing_tags-language-modeli","event_publishing_tags-large-language-model","event_publishing_tags-large-language-models","event_publishing_tags-linguistics","event_publishing_tags-llm","event_publishing_tags-llms","event_publishing_tags-mind","event_publishing_tags-natural-language-processing","event_publishing_tags-nlp","event_publishing_tags-openai","event_publishing_tags-yapay","event_publishing_tags-yapay-zeka","event_publishing_tags-zeka","event_publishing_tags-zihin","kategori-yazi"],"acf":[],"_links":{"self":[{"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/blog_content\/2168","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/blog_content"}],"about":[{"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/types\/blog_content"}],"version-history":[{"count":0,"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/blog_content\/2168\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/media\/2169"}],"wp:attachment":[{"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/media?parent=2168"}],"wp:term":[{"taxonomy":"event_publishing_tags","embeddable":true,"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/event_publishing_tags?post=2168"},{"taxonomy":"kategori","embeddable":true,"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/kategori?post=2168"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}